This study proposes a grasp pose detection algorithm to overcome key limitations of deep learning-based methods, such as data annotation demands and long training times. Using point cloud data, the algorithm identifies optimal grasp positions through geometric feature analysis and filters them based on stability and feasibility. It consists of four steps: point cloud acquisition, uniform sampling, candidate filtering via geometric features, and scoring-based selection. Experiments in virtual and real environments using a digital twin framework show average grasp success rates of 86.77% and 82.78%, demonstrating the method’s high efficiency and practical applicability.